geosimulation :: innovative geospatial simulation and analysis but innovative people

Home | Book | Research | Publications | Bio | Press | Geosimulation Labs
Dr. Paul M. Torrens, Center for Urban Science + Progress, New York University

Machine-learning behavioral geography

Project overview | Eye candy | Demo movie | Support | Related groups
Project overview

The goal of this project is to machine-learn behavioral rules for agent-based models, using data-mining and knowledge discovery on massive databases of trajectory samples from diverse sources. These data may come from location-aware hardware, such as Geographic Positioning Systems, alternative positioning systems (Wi-Fi, cell-phone triangulation), from geocoded trip diaries, or from observation. We are developing a scheme that can work with any of these data types, using only the simplest of geographical information: location in space and time.

This will allow us to build agent-based models for situations in which no theory exists, or to use machine-learning to better support theory-driven models by allying them to the real-world behavioral geography of actual people, in actual places, engaged in actual activities.

The scheme works as a combination of spatial database management, spatial data access, spatial analysis, classification, clustering, and weighted training. Initially, we are using data from a three-year observational study, for which we developed a customized space-time GIS observation and data-warehousing scheme.


Eye candy
The figure above illustrates sample trajectories from our three-year observational study, collected using a customized space-time GIS system that we developed to run on mobile hardware.
The figure above illustrates the machine-learned agent-based model running in real-time, constantly learning its path through space and time using only a library of trajectory samples and our knowledge discovery model.
nsf Torrens, P.M; Ghanem, Roger; Kevrekidis, Yannis (2010-2011). "Accelerating innovation in agent-based simulations: Application to complex socio-behavioral phenomena". National Science Foundation (Division of Civil and Mechanical Systems)
Torrens, P.M. (2007-2012) “CAREER: Exploring the dynamics of individual pedestrian and crowd behavior in dense urban settings: a computational approach”. National Science Foundation (Faculty Early Career Development (CAREER); Geography & Regional Science/ Methodology, Measurement, and Statistics)
Related groups
GAMMA group at University of North Carolina, Chapel Hill


GIS movement tracks

Big data movement analytics


climate indicators spatial analysis

Land indicators of climate

geosimulation high performance computing

High-performance computing and networking for geosimulation

earthquake model agent based GIS

Earthquake models

CA ice sheet model

Ice-sheet modeling

kinect control of GIS and robots
Robot motion control

simulating disasters ABM GIS
Human behavior in critical scenarios

crowd model riot model simulation wired

Modeling riots

physics engine GIS

Dynamic physics for built infrastructure

moving agents through space and time

Moving agents through space and time

validating agent based models

Validating agent-based models

machine learning GIS

Machine-learning behavioral geography

high performance computing urban simulation emergence

Accelerating agent-based models

megacity models

Megacity futures

immersive modeling

Immersive modeling

space-time GIS

Space-time GIS and analysis

measuring sprawl

A toolkit for measuring sprawl

space-time GIS

Modeling time, space, and behavior

simulating crowd behavior

Simulating crowd behavior

wi-fi geography

Wi-Fi geography

Simulating sprawl

Simulating sprawl